Harendra Kumar
d4363a8ad7
Add loop tests
2017-08-28 02:09:08 +05:30
Harendra Kumar
86f575d6de
simplify pullworker code
2017-08-28 00:40:17 +05:30
Harendra Kumar
5db931586e
Simplify push code
2017-08-28 00:23:14 +05:30
Harendra Kumar
2cc5c120ac
Move a test code block
2017-08-27 08:08:48 +05:30
Harendra Kumar
064bb7e2cd
Some simplifications
2017-08-27 08:08:18 +05:30
Harendra Kumar
bdaefb3cff
Reuse threads instead of creating fresh every time
...
Improved the empty thread benchmark by more than 5 times. Work is now queued
to a common queue and all the worker threads pick it up from there.
2017-08-27 02:23:21 +05:30
Harendra Kumar
e8a746382d
Use TBQueue instead of an unbounded channel
2017-08-26 03:03:45 +05:30
Harendra Kumar
b3faafaecb
Solve the problem of last dispatch blocking drain
...
By always having the last dispatch in another thread. This has significant
impact on performance due to an extra thread or some other reason?
2017-08-25 23:58:38 +05:30
Harendra Kumar
2248afac5e
Simplify runner functions
2017-08-25 23:38:54 +05:30
Harendra Kumar
ab91267967
Add test case for hierarchical alternative composition
2017-08-25 23:38:06 +05:30
Harendra Kumar
5181c0d15f
Fix warning in benchmark
2017-08-24 03:49:32 +05:30
Harendra Kumar
1f01190cca
Fix bug due to typo in pull finalizer and move it
2017-08-24 03:46:44 +05:30
Harendra Kumar
a61b30ba71
Fix, delete, move examples
2017-08-24 03:45:29 +05:30
Harendra Kumar
02a745f8e9
update examples README
2017-08-24 03:08:06 +05:30
Harendra Kumar
7beefcec67
Fix examples
2017-08-24 03:07:14 +05:30
Harendra Kumar
b68840b6e6
Add some more examples
2017-08-24 02:22:49 +05:30
Harendra Kumar
5349a44ef3
update some examples
2017-08-24 02:16:02 +05:30
Harendra Kumar
d8c399e567
Implement tracking and killing of child threads
2017-08-22 21:45:58 +05:30
Harendra Kumar
36ecdf82e3
Perf - do not save pending work in the context
2017-08-22 17:17:35 +05:30
Harendra Kumar
6ec80a0f92
Fix draining of threads (serial instead of parallel)
2017-08-22 17:01:07 +05:30
Harendra Kumar
04fef4fe4f
Move a function in dep order
2017-08-21 22:31:42 +05:30
Harendra Kumar
09283c42ac
Implement pull based concurrency rate control
2017-08-21 21:10:09 +05:30
Harendra Kumar
b1ee90ec4e
Add the yield API
2017-08-19 03:22:59 +05:30
Harendra Kumar
500d889859
Simplify push-pull split, fix monoid instance
2017-08-18 07:12:29 +05:30
Harendra Kumar
b2faaf91ba
remove a couple of stale comments
2017-08-17 19:18:22 +05:30
Harendra Kumar
b38be3a834
Reduce channel communication for short tasks
...
Combine the last done message with yield
2017-08-17 19:07:37 +05:30
Harendra Kumar
91a6cee5ed
Simplify push
2017-08-17 18:14:44 +05:30
Harendra Kumar
9b2ca30312
Add the "for" parallel API and change "each" to serial
2017-08-17 04:42:45 +05:30
Harendra Kumar
63f965a784
cleanup Monoid instance
2017-08-17 01:43:14 +05:30
Harendra Kumar
6ea73b3d10
Cleanup and refactor threading stuff
2017-08-17 01:29:56 +05:30
Harendra Kumar
e874ecf575
Remove quadratic complexity due to serially-in-serially
2017-08-16 15:18:17 +05:30
Harendra Kumar
8a1d5d995e
Add a test for "each" to test multiple threads
2017-08-16 14:53:30 +05:30
Harendra Kumar
550eed458f
Make monoid instance append two computations in order
...
Conflicts:
test/Main.hs
2017-08-15 18:41:20 +05:30
Harendra Kumar
ef006c846d
Implement fork for Alternative instance
2017-08-15 18:36:26 +05:30
Harendra Kumar
9f7d672422
Simplify toList runner
2017-08-11 13:36:07 +05:30
Harendra Kumar
c9b9ab35db
Pass context directly, removing the use of State monad
2017-08-11 12:59:33 +05:30
Harendra Kumar
6935930cae
Add list-t to benchmarks
2017-08-10 15:25:28 +05:30
Harendra Kumar
f4cce2f9a3
Separate the bind and <|> semantics for parallelism
2017-08-10 15:24:47 +05:30
Harendra Kumar
7b52ccfff6
Partially fix the recording stuff
...
handlResult is still to be used
2017-08-10 00:13:05 +05:30
Harendra Kumar
f8851aa158
Add MonadRecorder instance to AsynclyT
2017-08-09 18:21:28 +05:30
Harendra Kumar
4f3775d51f
Enable threads API and tests
2017-08-09 17:44:05 +05:30
Harendra Kumar
17b30bafb7
rename the runner APIs
2017-08-09 17:30:33 +05:30
Harendra Kumar
504800bebe
More cleanup in RunAsync
2017-08-09 17:19:02 +05:30
Harendra Kumar
bb5a74d32a
Some cleanups
2017-08-09 16:22:25 +05:30
Harendra Kumar
eca7d81223
Remove redundant composition primitives
...
These can be simulated using a wait on an event and lifting/discarding
the result.
2017-08-09 15:37:10 +05:30
Harendra Kumar
a20c180451
Some cleanup
2017-08-09 02:11:16 +05:30
Harendra Kumar
99d9ba6c07
Fix the composite benchmark code
2017-08-08 20:37:11 +05:30
Harendra Kumar
c1f0a867c3
Continuation based implementation
...
Seems to have similar (slightly worse) performance as the lazy impl without
continuations.
lazy-streaming witout continuations without StateT = 400 ms
with continuations without StateT = 550 ms
However with continuations we do not need SPECIALIZE which is a significant
advantage to freely use in any Monad wihtout worrying about perf.
2017-08-08 15:23:14 +05:30
Harendra Kumar
7e8209e2d6
Separate the State layer from AsyncT layer
...
Performance degrades by 2x
StateT/AsyncT combined = 560 ms
Separated AsyncT (StateT Context IO) = 1.2 s
AsyncT IO in the separated version = 400 ms
2017-08-07 19:55:38 +05:30
Harendra Kumar
4d7024fae0
Initial poc of lazy streaming
2017-08-07 15:15:09 +05:30
